Does Decapitation Work? Does Decapitation Patrick B

Home , Decapitation, Robert Pape, Shaul Mishal

Does Decapitation Work? Does Decapitation Patrick B. Johnston Work? Assessing the Effectiveness of Leadership Targeting in Counterinsurgency Campaigns

Targeting of militant leaders is central to many states’ national security strategies, but does it work? What should policymakers expect when armed forces capture or kill militant leaders? Is leadership decapitation more likely to succeed or fail under certain conditions? These questions have never been more pressing than after the May 2011 killing of al-Qaida leader Osama bin Laden. As relevant as these questions are to current U.S. policy and strategy, they are also fundamental questions of asymmetric warfare. They matter because almost all policies of “high-value” targeting require difªcult judgments concerning both the potential consequences and the opportunity costs of targeting militant leaders. The decision to target enemy leaders requires that policymakers adjudicate among numerous difªcult, and potentially contradictory, choices. Leadership targeting strategies affect how states allocate scarce military, intelligence, and economic resources; how they construct their counterinsurgency or counterterrorism postures; and how interested foreign and domestic audiences react to their behavior. Despite the stakes, scholars have shown relatively little interest in leadership decapitation. Those who have written on the topic have tended to argue that leadership targeting is ineffectual and can be counterproductive.1 Tar-

Patrick B. Johnston is Associate Political Scientist at the RAND Corporation. He wrote this article while he was a fellow at the Center for International Security and Cooperation and the Empirical Studies of Conºict Project at Stanford University and at the International Security Program at Harvard Kennedy School’s Belfer Center for Science and International Affairs. A supplemental appendix is available online at http:// patrickjohnston.info.

The author would like to thank Max Abrahms, Mia Bloom, Stephanie Carvin, Martha Crenshaw, Lynn Eden, James Fearon, Joseph Felter, Bryan Price, William Reno, Anoop Sarbahi, Paul Staniland, David Steinberg, and the anonymous reviewers for commenting on earlier versions of this article.

1. Robert A. Pape, Bombing to Win: Air Power and Coercion in War (Ithaca, N.Y.: Cornell University Press, 1996); Robert A. Pape, “The Strategic Logic of Suicide Terrorism,” American Political Science Review, Vol. 97, No. 3 (August 2003), pp. 1–19; Stephen T. Hosmer, Operations against Enemy Leaders (Santa Monica, Calif.: RAND, 2001); Paul Staniland, “Defeating Transnational Insurgencies: The Best Offense Is a Good Fence,” Washington Quarterly, Vol. 29, No. 1 (Winter 2005/06), pp. 21–40; Mohammed M. Hafez and Joseph M. Hatªeld, “Do Targeted Assassinations Work? A Multivariate Analysis of Israel’s Controversial Tactic during Al-Aqsa Uprising,” Studies in Conºict and Terrorism, Vol. 29, No. 4 (June 2006), pp. 359–382; Or Honig, “Explaining Israel’s Misuse of Strategic Assassi-

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/ISEC_a_00076 by guest on 26 September 2021 International Security 36:4 48

geting enemy leaders, argues Robert Pape, “has never been effective” as a coercive tool in interstate war.2 Decapitating terrorist organizations is not effective either, according to Pape, whose high-proªle study of suicide terrorism concludes that decapitation strategies “have met with meager success.”3 Jenna Jordan’s recent study supports Pape’s earlier conclusion. Terrorist organizations rarely collapse after their top leaders are captured or killed, she ªnds. Jordan concludes that leadership decapitation is “a misguided strategy.”4 This consensus is premature. Researchers have conducted few systematic as- sessments of leadership decapitation’s effectiveness; evidence remains scant.5 But contrary to scholars’ claims that leadership decapitation never works, the evidence appears to be more mixed. In numerous cases, decapitation was vital in degrading and defeating militant groups. In Peru, for example, Shining Path leader Abimeal Guzmán’s 1992 capture crippled the group’s bid for power. In Turkey, the capture of Abdullah Ocalan, leader of the Kurdistan People’s Party, in 1999 precipitated the group’s steep decline. And in Italy, authorities used critical intelligence obtained from captured Red Brigades leaders to dismantle remnants of the organization.6 Moreover, the research design and methodologies used in nearly all previous studies make it difªcult to draw credible conclusions about the impact of leadership decapitation. Three problems are common in the literature. First, the security studies literature, for example, relies uniformly on no-variance designs.7 Yet credible causal inferences cannot be made from studies that only

nations,” Studies in Conºict and Terrorism, Vol. 30, No. 6 (June 2007), pp. 563–577; and Jenna Jordan, “When Heads Roll: Assessing the Effectiveness of Leadership Decapitation,” Security Studies, Vol. 18, No. 4 (December 2009), pp. 719–755. At least one study suggests that decapitation is effective. See Bryan Price, “Removing the Devil You Know: Unraveling the Puzzle behind Decapitation Effectiveness and Terrorist Group Duration,” Ph.D. dissertation, Stanford University, 2009. Several authors suggest that decapitation’s impact is conditional on other factors. See, for example, Daniel Byman, “Do Targeted Killings Work?” Foreign Affairs, Vol. 85, No. 2 (April 2006), pp. 95–111; Daniel Byman, A High Price: The Triumphs and Failures of Israeli Counterterrorism (New York: Oxford Uni- versity Press, 2011); Matt Frankel, “The ABCs of HVT: Key Lessons from High Value Targeting Campaigns against Insurgents and Terrorists,” Studies in Conºict and Terrorism, Vol. 34, No. 1 (Janu- ary 2010), pp. 17–30; and Austin Long, “Assessing the Success of Leadership Targeting,” CTC Sen- tinel, Vol. 3, Nos. 11/12 (November 2010), pp. 19–21. 2. Pape, Bombing to Win, p. 316. 3. Pape, “The Strategic Logic of Suicide Terrorism,” p. 14. 4. Jordan, “When Heads Roll,” p. 754. 5. Few large-n studies of leadership decapitation exist. See Jordan, “When Heads Roll”; Price, “Re- moving the Devil You Know”; and Patrick Johnston, “The Effectiveness of Leadership Decapita- tion in Counterinsurgency,” paper presented at CISAC Social Science Seminar, Stanford University, May 2009. 6. In addition, Aaron Mannes’s medium-n study found “some indication that decapitation strikes can be effective in reducing terrorist group incidents.” Mannes’s results, however, were not statistically signiªcant. Mannes, “Testing the Snake Head Theory: Does Killing or Capturing Its Leaders Reduce a Terrorist Group’s Activity?” Journal of International Policy Solutions, Vol. 9 (Spring 2008), pp. 43–44. 7. A notable exception, published in the economics literature, is Benjamin Jones and Benjamin

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/ISEC_a_00076 by guest on 26 September 2021 Does Decapitation Work? 49

examine cases in which opposing leaders were captured or killed. Second, scholars have tended to use extremely restrictive coding criteria, setting the bar unrealistically high for decapitation to be considered successful; except in cases where the target was quickly and decisively defeated following a leader’s capture or death, scholars have usually coded decapitation as a failure. This is appropriate for evaluating decapitation’s proximate strategic impact, but it is inappropriate for assessing decapitation’s longer-term political, military, and economic effects. Decapitation could have a wide variety of effects—some positive, others potentially negative—that this approach does not capture.8 A second problem is selection bias. Militant leaders tend to be killed or captured at key junctures in campaigns—periods when governments may already be more likely to win or lose.9 Security studies scholarship on leadership decapitation—including large-N studies—fails to address this issue, which makes it difªcult to identify whether decapitation explains the outcomes of interest, or whether other factors that make decapitation more likely to occur actually drive the relationship. This challenge is daunting, both for quantitative and qualitative studies that rely on observational data. An experimental design would solve this problem, but for many of the most important security studies questions, especially the present one, an experiment is neither desirable nor feasible. In these situations, scholars can instead exploit research designs that focus on data where confounding factors are unlikely to cause misleading correlations. Doing so helps scholars to isolate their variable(s) of interest and to avoid making conclusions and policy recommendations based on spurious, misleading ªndings. This article addresses these challenges by analyzing a large number of cases in which governments attempted—successfully and unsuccessfully—to remove top militant leaders and the events that followed these attempts.10 This approach relies on a natural experiment to help isolate the importance of militant leaders. To the extent that chance plays a role in the outcome of operations aimed at removing military leaders, this approach can help researchers assess the likelihood that organizations whose leaders were captured or killed would have fared differently had their leadership remained intact.

Olken, “Hit or Miss? The Effect of Assassinations on Institutions and War,” American Economic Journal: Macroeconomics, Vol. 1, No. 2 (July 2009), pp. 55–87. 8. This hypothesis is key to policy and doctrine, which prescribe leadership targeting as part of a general strategy to reduce the threat of militant attacks by degrading the organizational capacity of militants to conduct large-scale operations. 9. James D. Fearon and David D. Laitin, “Civil War Termination,” unpublished manuscript, Stan- ford University, 2008, pp. 39–42; and Jones and Olken, “Hit or Miss?” 10. For a similar approach to studying the effects of national leaders on economic growth, democ- ratization, and war, see Benjamin Jones and Benjamin Olken, “Do Leaders Matter? National Lead- ership and Growth since World War II,” Quarterly Journal of Economics, Vol. 120, No. 3 (August 2005), pp. 835–864; and Jones and Olken, “Hit or Miss?”

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/ISEC_a_00076 by guest on 26 September 2021 International Security 36:4 50

This approach is valid for identifying the effect of leadership decapitation as long as decapitation operations’ outcomes are not systematically determined by factors unaccounted for in the analysis. This assumption holds up to both anecdotal and systematic scrutiny. As I argue below, the clandestine nature of militant leadership and unforeseen or idiosyncratic circumstances can com- promise the chances of even the most well planned operations to succeed. Large-N scrutiny supports this argument. Analysis of 118 decapitation attempts from a sample of 90 counterinsurgency campaigns shows that factors commonly associated with counterinsurgency success generally fail to predict the success or failure of government actions to remove militant leaders. My results challenge previous claims that removing militant leaders is ineffective or counterproductive. On the contrary, they suggest that leadership decapitation (1) increases the chances of war termination; (2) increases the probability of government victory; (3) reduces the intensity of militant violence; and (4) reduces the frequency of insurgent attacks. It is important that the ªndings be interpreted with care and in light of their limitations. Although the estimated impact of decapitation is substantial, decapitation is not a silver bullet. A better interpretation is that, although decapitation is likely to help states’ overall efforts against militant organizations, other factors will also matter greatly in most cases.11 In other words, decapitation is more likely to help states achieve their objectives as an operational com- ponent within an integrated campaign strategy than as a stand-alone strategy against insurgent and terrorist organizations. The article proceeds as follows. First, I describe and critique the argument that leadership decapitation is ineffective. Second, I describe my dataset, variables, and key summary statistics. Third, I discuss my empirical methodology. I begin by presenting evidence that decapitation attempt outcomes are plausibly exogenous. Then I describe my estimation strategy for identifying the effects of leadership decapitation. Fourth, I present the main results of my analysis. I conclude the article by recapitulating my primary ªndings and dis- cussing their policy implications.

The Notion That Leadership Decapitation Does Not Work

Claims that leadership decapitation is ineffective or counterproductive usually rest on three premises: organizational durability, martyrdom effects, and decentralizing effects. I examine each premise in turn.

11. Matt Frankel emphasizes this point in his research on high-value targeting. Frankel, “The ABCs of HVT,” pp. 20, 28. For an in-depth example of how integrating leadership targeting with

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/ISEC_a_00076 by guest on 26 September 2021 Does Decapitation Work? 51

organizational durability Individuals—including leaders—rarely factor signiªcantly into leading theories of insurgency and counterinsurgency, which instead emphasize the civilian population’s importance.12 These theories suggest that a multifaceted array of actors, institutions, and structures, which are rooted deeply in society, make up insurgent organizations. Such social organizations are likely to be highly durable and larger than any individual or set of individuals. It follows from the theories, then, that removing key individuals is unlikely to undermine insurgency; effectively undermining insurgency requires the removal of societal support. Assuming that insurgent organizations are resilient to the loss of individual leaders holds intuitive appeal, especially when considering high-proªle movements such as the National Liberation Front in Algeria or the Vietcong, but the assumption’s applicability to the wider universe of insurgencies is question- able. It is not unreasonable to think that many guerrilla groups would lack the robust structures necessary to insulate themselves from the shocks associated with the loss of key leaders. The Janatha Vimukthi Peramuna (JVP), for example, a communist insurgency that staged uprisings against the Sri Lankan government in 1971 and 1987, suffered greatly in each conºict after Sri Lankan authorities captured its leader, Rohana Wijeweera, who masterminded the group’s overall strategy and tactical operations. In both conºicts, the JVP re- lied heavily on Wijeweera’s operational skills and charisma. After his capture, Wijeweera’s subordinate commanders failed to execute the operations he had planned in each subsequent campaign, and the JVP was quickly defeated.13 Palestinian Islamic Jihad (PIJ) leader Fathi Shaqaqi was similarly crucial to his organization’s cohesion and operational capabilities.14 It is therefore unsur- prising that Shaqaqi’s 1995 assassination—allegedly by Israeli agents—dealt the PIJ a severe blow.15 Finally, even though scholars correctly observe that leadership decapitation is rarely a silver bullet solution to insurgency, this does not mean that killing or

other countermilitancy policies can enhance government campaign effectiveness, see Byman, “Do Targeted Killings Work?” pp. 105–106. 12. Mao Zedong, On Guerrilla Warfare (New York: Praeger, 1961); David Galula, Counterinsurgency Warfare: Theory and Practice (Westport, Conn.: Praeger Security International, 1964); and the U.S. Army and Marine Corps, The U.S. Army/Marine Corps Counterinsurgency Field Manual, FM 3-24 (Chicago, Ill.: University of Chicago Press, 2007). 13. S. Arasaratnam, “The Ceylon Insurrection of April 1971: Some Causes and Consequences,” Paciªc Affairs, Vol. 45, No. 3 (Autumn 1972), pp. 356–371. 14. Steven R. David, “Fatal Choices: Israel’s Policy of Targeted Killing,” in Efraim Inbar, ed., De- mocracies and Small Wars (London: Frank Cass, 2003), p. 6. 15. For more detail, see the organizational proªle of PIJ on the website of the National Consortium for the Study of Terrorism and Responses to Terrorism (START), http://www.start.umd.edu/ start/data_collections/tops/terrorist_organization_proªle.asp?idϭ82.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/ISEC_a_00076 by guest on 26 September 2021 International Security 36:4 52

capturing an insurgency’s leadership cannot diminish insurgencies’ organizational capabilities and effectiveness. On the contrary, removing key leaders might have important effects, such as degrading insurgencies’ pool of skilled commanders, strategists and operatives; disrupting insurgents’ planning, training, and execution of operations and attacks; and, by putting remaining insurgents on the defensive, assisting government forces in seizing or main- taining the tactical and operational initiative.

martyrdom effects A popular assumption is that costly “martyrdom effects” are a consequence of capturing or killing top militant leaders.16 Instead of degrading groups’ morale and esprit de corps, decapitation increases insurgents’ resolve, aids their recruitment efforts, and intensiªes their desire to use violence in retaliation for state actions.17 Jordan, for example, argues that “going after the leader may strengthen a terrorist group’s resolve, result in retaliatory attacks, increase public sympathy for the organization, or produce more lethal attacks.”18 While the commonly held view of insurgents and terrorists as irrational fa- natics makes the martyrdom argument seductive, there are two problems with it. First, it tacitly assumes that insurgent leaders are popular and that their removal will be unwelcome among subordinate members of the organization and the population at large. Yet it can be these very actors who covertly provide the intelligence used by governments to target insurgent or terrorist leaders.19 This fact suggests that militant leaders who are targeted may be less popular among key followers than the martyrdom theory would suggest. Second, the martyrdom argument rests on the premise that emotions are the primary driver of insurgent behavior. This claim clashes with leading scholarship on insurgency, which suggests that militants are likely to make difªcult choices on the basis of strategic calculation rather than on emotion alone.20

16. Byman, “Do Targeted Killings Work?” pp. 99–100; and David, “Fatal Choices,” pp. 8–9. 17. Catherine Lotrionte, “When to Target Leaders,” Washington Quarterly, Vol. 26, No. 3 (Summer 2003), pp. 73–86; Edward H. Kaplan, Alex Mintz, Shaul Mishal, and Claudio Samban, “What Hap- pened to Suicide Bombings in Israel? Insights from a Terror Stock Model,” Studies in Conºict and Terrorism, Vol. 28, No. 3 (August 2005), pp. 225–235; and Jordan, “When Heads Roll,” pp. 82–83. 18. Jordan, “When Heads Roll,” p. 755. 19. See Johnston, “The Effectiveness of Leadership Decapitation in Counterinsurgency,” p. 8. 20. On strategic decisionmaking in insurgency, see especially Stathis N. Kalyvas, The Logic of Vio- lence in Civil War (New York: Cambridge University Press, 2006); Jeremy M. Weinstein, Inside Rebel- lion: The Political Economy of Rebel Organization (New York: Cambridge University Press, 2006); Roger D. Petersen, Understanding Ethnic Violence: Fear, Hatred, and Resentment in Twentieth-Century Eastern Europe (New York: Cambridge University Press, 2002); Elisabeth J. Wood, Insurgent Collec- tive Action and Civil War in El Salvador (New York: Cambridge University Press, 2003); and Rogers Brubaker and David D. Laitin, “Ethnic and Nationalist Violence,” Annual Review of Sociology, Vol. 24 (August 1998), pp. 423–452.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/ISEC_a_00076 by guest on 26 September 2021 Does Decapitation Work? 53

So while the killing or capture of a guerrilla leader can lead to public out- cries from his supporters, it can also signal to the insurgency’s potential re- cruits and ascendant leadership that the incumbent possesses signiªcant strength and intelligence capabilities, possibly deterring additional mobiliza- tion for insurgency.21

inadvertent decentralization Third, analysts have suggested that leadership decapitation makes defeating insurgencies more difªcult by decentralizing them.22 The assumption is that, as organizations become less hierarchical, they evolve into localized cells capa- ble of persisting long after other parts of the organization have been neutral- ized.23 Yet there is little empirical evidence that decapitation ºattens militant organizational structures or that decentralized groups are more resilient to state countermeasures. On the contrary, the opposite hypothesis is plausi- ble: even if decapitation does induce decentralized militant organization— characterized by small cells with limited contact or knowledge of one another—the net effect of these dynamics may be a weakening of the organization that results from the difªculty or insecurity of movement and com- munications among various cells, which makes organizing the large-scale collective action necessary to conduct persistent, sophisticated attacks much more difªcult. In the next section, I discuss the dataset and variables used in the empirical analysis.

Data

This article focuses on attempted removals of insurgent leaders in counterinsurgency campaigns, where the “leader” is deªned as the most powerful ªgure or ªgures in an insurgent organization. My focus is necessarily limited to top leaders; systematically identifying and distinguishing among upper-

21. Alexandre S. Wilner, “Targeted Killings in Afghanistan: Measuring Coercion and Deterrence in Counterterrorism and Counterinsurgency,” Studies in Conºict and Terrorism, Vol. 33, No. 4 (April 2010), pp. 307–329. 22. For examples of this dynamic, see Byman, “Do Targeted Killings Work?” pp. 100–101; and Da- vid, “Fatal Choices,” p. 8. 23. On this general dynamic in terrorist organizations and its implications for counterterrorism, see Marc Sageman, Leaderless Jihad: Terror Networks in the Twenty-First Century (Philadelphia: Uni- versity of Pennsylvania Press, 2008). See also Bruce Hoffman, “The Myth of Grass-Roots Terror- ism,” Foreign Affairs, Vol. 87, No. 3 (May/June 2008), pp. 133–138. On organizational forms and leadership decapitation in counterterrorism, see Audrey Kurth Cronin, “How al-Qaida Ends: The Decline and Demise of Terrorist Groups,” International Security, Vol. 31, No. 1 (Summer 2006), pp. 7–48; and Jordan, “When Heads Roll.”

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/ISEC_a_00076 by guest on 26 September 2021 International Security 36:4 54

echelon, midtier, and low-level leaders in clandestine organizations is prohibi- tively difªcult. By contrast, the relative visibility of top leaders enables their identiªcation in a way that minimizes measurement error that could lead to misleading results. Focusing on top leaders is further justiªed by practical as- pects of my research question. Because top leaders are the most important members of insurgent organizations and the highest-priority targets of anti- insurgent operations, the most urgent ªrst-order issue, for both scholars and policymakers, is to identify whether removing these speciªc individuals mat- ters for key outcomes of interest in counterinsurgency.24

dataset To collect the data necessary for this study, I compiled a list of insurgent leaders from the START Terrorist Organization Proªles (TOPS) database at the University of Maryland.25 Having identiªed the top group leader or leaders, I then collected data on leadership decapitation from the Lexis-Nexis Academic Universe database.26 I performed keyword searches on the entire collection of English-language news sources and reviewed the results for information on leadership removals and attempted removals. I cross-validated each attempt identiªed in the search results with at least one additional source before tag- ging it for inclusion in the dataset. I restricted my data collection to plausibly exogenous attempts to kill or capture insurgent leaders. Examples of plausibly exogenous decapitation events include attempts to remove leaders through assassination plots in which bombs are planted or shots are ªred; combat operations in which ªreªghts or air strikes directly target leaders’ units; and raids or sweeps of leaders’ com- pounds or camp areas. Rumored plots that never materialized, and operations in which leaders escaped before an attempt was made, are not plausibly exogenous and were excluded. For each attempt that satisªed these criteria, I coded the date, location, attempt type, and outcome. After ªltering the data through these criteria, I was able to document 118 attempts. Forty-six of the 118 attempts (39%) resulted in the removal of a top-level insurgent leader. A potential concern is that failed attempts are underrepresented in the

24. Future research is necessary to address second-order questions about the impact of removing upper-echelon and midtier insurgents and terrorists from militant organizations. Targeting the middle ranks of insurgencies has indeed been key to Israeli counterterrorism strategy as well as to U.S. operations in Afghanistan and Iraq. 25. The TOPS database contains information on the top leader or leaders of most of the insurgencies in my sample. I coded group leaders who were not identiªed in the TOPS database using historical encyclopedias and secondary sources. See the database online at http://www.start.umd .edu/start/data_collections/tops/. 26. The Lexis-Nexis website is accessible at http://www.lexisnexis.com/.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/ISEC_a_00076 by guest on 26 September 2021 Does Decapitation Work? 55

dataset. One might expect, for example, that failed attempts are underreported because concealing them might be in the government’s interest. Upon closer examination, however, theoretical and methodological considerations mitigate this issue. Theoretically, it is likely that insurgent groups would have a similar interest in revealing governments’ unsuccessful attempts. Uncovering failed attempts might, for example, feed into insurgent propaganda by demonstrat- ing the government’s incompetence and unscrupulous behavior while making insurgent leaders appear cleverer than government forces and occupants of the moral high ground. As a result, even if governments would prefer to con- ceal unsuccessful attempts, countervailing messaging by insurgents is likely to thwart their efforts. Methodologically, even if instances of failure are missing from the dataset, it would reduce the precision of my regression estimates’ standard errors but not undermine causal identiªcation unless a narrow set of unlikely circumstances were to obtain—namely, if the likelihood of failing to observe decapitation attempts was systematically higher around periods when changes in the dependent variable are observed. This is unlikely given that (1) the data are drawn from news reports, and (2) media reporting tends to be most intense during critical periods of campaigns captured by my dependent variables. In fact, it is possible that if any reporting bias does exist, it occurs during periods when change is not observed on the dependent variables. This would bias my estimates downward, not upward. Another consideration is whether reporting differs by regime type given variations in press freedom. A “counterinsurgency campaign” refers to a military contest between a state and a nonstate actor or actors that employed a strategy of guerrilla warfare. Following leading scholarship on insurgency and counterinsurgency, I deªne “guerrilla warfare” as a strategy of armed resistance in which small, mobile bands inºict violence using unconventional military tactics while avoiding direct battle when possible and operating among the civilian population.27 I used four criteria to identify appropriate campaigns for analysis. The ªrst is asymmetric conventional capabilities. Power asymmetries can be observed when the relative sophistication and lethality of counterinsurgent forces’

27. See, for example, James D. Fearon and David D. Laitin, “Ethnicity, Insurgency, and Civil War,” American Political Science Review, Vol. 97, No. 1 (February 2003), pp. 75–90; Benjamin Valentino, Paul Huth, and Dylan Balch-Lindsay, “‘Draining the Sea’: Mass Killing and Guerrilla Warfare,” In- ternational Organization, Vol. 58, No. 2 (April 2004), pp. 375–407; Weinstein, Inside Rebellion; Jason Lyall and Isaiah Wilson III, “Rage against the Machines: Explaining Outcomes in Counterinsur- gency Wars,” International Organization, Vol. 63, No. 1 (Winter 2009) pp. 67–106; Jason Lyall, “Do Democracies Make Inferior Counterinsurgents? Reassessing Democracy’s Impact on War Out- comes and Duration,” International Organization, Vol. 64, No. 1 (Winter 2010), pp. 10–16; and Stathis N. Kalyvas and Laia Balcells, “International System and Technologies of Rebellion: How the End of the Cold War Shaped Internal Conºict,” American Political Science Review, Vol. 104, No. 3 (August 2010), pp. 415–429.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/ISEC_a_00076 by guest on 26 September 2021 International Security 36:4 56

weapons or other war-ªghting technologies are more sophisticated than those used by insurgents. Second, insurgents using guerrilla warfare typically oper- ate within the civilian population rather than on a battleªeld separate from the populace. Although it is difªcult to discern whether insurgents actually seek uncoerced popular support, observing whether militant organizations implant themselves in the civilian population is easier. The third criterion is tactical mismatch. Unlike the raison d’être of the regular armies dispatched to sup- press them, insurgencies rely on unconventional military tactics, such as am- bushes, hit-and-run attacks, and sabotage to pursue their objectives. Militants must be observed using these tactics while generally avoiding direct battle to qualify for inclusion in the dataset. The ªnal operational criterion is a one- month minimum duration rule. These criteria help ensure that all campaigns in the dataset were sustained, asymmetric violent conºicts between organized military actors, not brief, disorganized bursts of violence or instances of riot- ing. Campaigns that failed to satisfy each criterion were excluded from the analysis. In all, I identiªed 90 campaigns from 1975 to 2003 that satisªed these criteria.28 In total, the dataset contains 928 campaign-year observations, a suf- ªcient number to detect statistical effects. To estimate the effect of leadership decapitation on counterinsurgency campaign outcomes and dynamics, I used two types of measures. The ªrst involves campaign outcomes, which come from the Correlates of War (COW) project.29 I examined campaign outcomes in two ways: termination and success. Termination is a dummy variable, where “1” indicates that the campaign dropped below a minimal threshold of violence in a given campaign-year. Success is also measured as a dummy variable, where “1” indicates that a campaign ended on terms favorable to the counterinsurgent in a given campaign- year and “0” indicates that it did not. The second type of measure involves conºict dynamics. These data come from the Global Terrorism Database.30 I also examined conºict dynamics in two ways: conºict intensity and insurgent-initiated incidents. Conºict intensity measures the total number of conªrmed fatalities, by group, in each

28. My analysis is limited to campaigns from 1975 to 2003 because extending the analysis as far back as 1945—the standard start date of several leading civil wars datasets—would likely produce signiªcant undercounts of the “failure” variable, as less public information is available prior to the mid-1970s, when an expanded global news media and new restrictions on secrecy and covert operations led to more reporting of state attempts to remove leaders of both states and insurgent organizations. 29. For campaigns not in the COW project data, I consulted Jason Lyall and Isaiah Wilson III, “Correlates of Insurgency” dataset, 2009; the Global Terrorism Database; and other secondary sources. 30. For full documentation, see the GTD codebook, p. 21, http://www.start.umd.edu/gtd/ downloads/Codebook.pdf.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/ISEC_a_00076 by guest on 26 September 2021 Does Decapitation Work? 57

Table 1. Summary Statistics: Decapitation Atttempts

Percentage Leader Removed Type Observations (total attempts) (percentage successful)

Shooting 34 29% 44% Bombing 17 14% 29% Raid/Sweep 25 21% 64% Combat 41 34% 32% Unknown 3 2% 0% Total attempts 119 100% 40%

campaign-year. This variable proxies overall conºict violence and insurgent lethality.31 Insurgent-initiated incidents measures the number of attacks and other violent incidents, by group, aggregated to the campaign-year level.32 This variable proxies insurgent groups’ levels of activity and operational tem- pos. Taken together, these variables enabled me to test how leadership decapitation affects both the lethality and rate of insurgent activities.

summary statistics Table 1 provides summary statistics about decapitation attempts. With regard to the manner in which attempts have been conducted, combat operations have been the most common, comprising 34 percent of attempts, and shootings the second most common, used in 29 percent of attempts. Attempts that occurred in combat operations succeeded at removing insurgent leaders in 32 percent of the cases, and shooting attempts succeeded in 44 percent of the cases. Raids and sweep operations have the highest success rate, at 64 percent, and attempts to kill leaders using bombs and other explosive devices appear to be particularly ineffective, as they succeeded in only 29 percent of the cases. Full summary statistics can be found in the online appendix.33

31. Because this variable includes government, civilian, and insurgent deaths, it is not a direct measure of insurgent violence. Although one should expect the conºict intensity variable to be noisy, it should still provide insight into whether removing insurgent leaders has a causal effect on conºict escalation or de-escalation because successful and failed attempts should not be correlated with observed or unobserved variables. 32. See the GTD codebook, p. 42. 33. The summary statistics include probabilities of transition from ongoing conºict to war termination and ongoing conºict to government victory. Assuming no temporal dependence, the probability of a campaign ending in a given campaign-year is about 9 percent. The probability of government winning in a given campaign-year is about 5 percent. For violence, the average number of people killed per campaign-year in my sample is approximately 66, and the average number of annual insurgent attacks is 16.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/ISEC_a_00076 by guest on 26 September 2021 International Security 36:4 58

Figure 1. Attempted and Successful Decapitation Strikes

Figure 1 shows how the frequency of decapitation events has evolved over time, plotting the frequency of attempts, and successful attempts, in each decade since the 1970s by campaign-year. The ªgure indicates that both the rate of leadership decapitation attempts and of successes increased around 2000. This increase coincided with escalated high-value targeting efforts after the terrorist attacks of September 11, 2001.34

Methodology

Identifying the effects of leadership decapitation is challenging. Establishing the direction of causation is particularly difªcult, because governments might target insurgent leaders disproportionately when they are either winning or losing. To address this challenge, I use a large set of attempts to remove insurgent leaders and exploit the “failures” as controls for the “successes.” My unit of analysis is thus the outcomes of decapitation attempts rather than leader-

34. On the renewed vigor with which governments have targeted militant leaders since the Sep- tember 11 attacks, see, in particular, Graham H. Turbiville Jr., Hunting Leadership Targets in Counter- insurgency and Counterterrorist Operations: Selected Perspectives and Experience, Joint Special Operations University, Report 07-6 (Tampa, Fla.: Joint Special Operations University Press, 2007), pp. 8–9.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/ISEC_a_00076 by guest on 26 September 2021 Does Decapitation Work? 59

ship decapitation itself. Although the timing of decapitation strikes is not plausibly exogenous—leaders are often targeted at key moments of campaigns when a government is more likely either to win or to lose—the data suggest that the success and failure of these attempts is, on average, exogenous.35 Accounts of “near misses” and “bad luck” are common in the historical re- cord. From the ªrst known assassination plot authorized by U.S. ofªcials against a nonstate actor, which targeted Pancho Villa during the 1916 Punitive Expedition, success in operations against high-value targets has been far from a foregone conclusion. In the case of Villa, the expedition sought to eliminate the legendary rebel for the threat he and his followers posed to American towns in southern Texas, New Mexico, and Arizona. The United States’ man- hunt extended deep inside Mexican territory, but Villa continually evaded U.S. forces. John Pershing, commander of the expedition, responded to these failures by recruiting a team to inªltrate Villa’s bands and kill him. The team man- aged to penetrate Villa’s camp and poison the rebel leader’s coffee during breakfast one morning. Villa reportedly drank at least half of the laced coffee but escaped with only minor illness.36 The United States’ failed attempt on Villa’s life foreshadowed the difªculties that states would face in their attempts to neutralize insurgent and terrorist leaders in the second half of the twentieth century. One example is the First Russo-Chechen War, in which the Russian government allegedly sponsored several failed assassination attempts against senior Chechen ofªcials. For in- stance, in July 1996, Russian intelligence ofªcials were informed that Chechen leader Dzhokar Dudayev would be chauffeured to an upcoming conference in Moscow by a driver named Khamad Kurbanov. Based on this information, Russian agents developed a plan to assassinate Dudayev during the Moscow trip. The plan was put into motion weeks before the assassination was to take place. It began when Russian police made what appeared to be a routine stop of Kurbanov’s vehicle at a Russian-manned checkpoint inside Chechnya. Kurbanov was brieºy taken inside a nearby ofªce for questioning. While the police questioned Kurbanov, a team of Russian agents planted explosives under the seats of his car. Once the explosives were in place, the Russians released Kurbanov from questioning. Kurbanov unwittingly drove the explosive-packed

35. The identifying assumption made in this type of research design is as follows: although attempts to remove leaders may be driven by historical circumstances, the success or failure of decapitation attempts can be treated as plausibly exogenous conditional on attempts taking place. See Jones and Olken, “Hit or Miss?” p. 56. 36. Pancho Villa was ªnally assassinated in 1923. For more on this attempt, see Stephen F. Knott, Secret and Sanctioned: Covert Operations and the American Presidency (New York: Oxford University, 1996), p. 171. See also Ward Thomas, “Norms and Security: The Case of International Assassina- tion,” International Security, Vol. 25, No. 1 (Summer 2000), p. 112.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/ISEC_a_00076 by guest on 26 September 2021 International Security 36:4 60

vehicle without incident for weeks before Dudayev’s Moscow trip. Mean- while, Russian operatives prepared to detonate the explosives once Dudayev’s presence in the vehicle could be conªrmed. Shortly before he was to ride with Kurbanov, however, Dudayev’s itinerary changed, and he was forced to ride with another driver. If not for this unexpected turn of events, Dudayev almost certainly would have been killed.37 As I demonstrate below, decapitation attempts fail more often than they succeed, and these outcomes are uncorrelated with key observable variables. As a result, failed attempts provide an ideal set of counterfactual observations that enable identiªcation of leadership decapitation’s causal impact.

exogeneity Using failed attempts as controls for successes assumes that attempt outcomes are uncorrelated with the error terms of the regression equations used to estimate the effects of leadership decapitation. I test this assumption by putting the independent variable on the left-hand side of the regression and examining whether variables in my dataset predict successful attempts. The results of this test are displayed in table 2. In column 1, the mean values of the variables are presented. These values are all taken from the year before decapitation strikes took place. Column 2 displays means of these variables in the year before failed attempts; column 3 displays the differences in the means of successful and unsuccessful decapitation attempts; and column 4 presents the results of two-sided t-tests of the equality of these means.38 Table 2 shows that the sample of successful and failed assassination attempts is indeed balanced across key variables: regime type; the counterinsurgent state’s logged gross domestic product per capita; the logged population in thousands of the counterinsurgent state; the logged average elevation of the conºict theater in meters; and the logged distance from the counterinsurgent’s capital city to the conºict theater in kilometers. The only variable for which the mean difference is statistically signiªcant is the counterinsurgent nation’s total population. The difference in means is signiªcant at the 10 percent level (p-value ϭ 0.07). This preliminary analysis demonstrates that the possibility that population is a confounding variable cannot be rejected.39

37. Gail Lapidus, “Contested Sovereignty: The Tragedy of Chechnya,” International Security, Vol. 23, No. 1 (Summer 1998), p. 18; and Svante Cornell, Small Nations and Great Powers: A Study of Ethnopolitical Conºict in the Caucasus (Richmond, U.K.: Curzon, 2000), p. 209. 38. These are results of t-tests that do not assume equal variance. 39. This evidence, however, is inconclusive; given that ªve variables are examined, it is not sur- prising that two speciªcations were statistically signiªcant. See the supplemental table 3 in the online appendix for additional analysis of sample balance. This analysis uses probit regressions of attempt outcome on the variables examined in table 2 instead of t-tests. This analysis suggests that

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/ISEC_a_00076 by guest on 26 September 2021 Does Decapitation Work? 61

Table 2. Are Successful and Failed Attempts Similar? Pairwise t-Tests of Sample Balance

Success Failure Difference p-Value Democracy ( 0.59 (0.53 0.05 0.42 (0.05) (0.04) (0.07) Gross domestic product per capita ( 7.69 (7.7 Ϫ0.01 0.95 (0.13) (0.15) (0.2) Population (11.13 (10.36 0.77 0.07 (0.32) (0.27) (0.42) Military personnel ( 5.24 (4.95 0.29 0.45 (0.29) (0.25) (0.38) Elevation (5.87 (6.09 Ϫ0.23 0.41 (0.24) (0.14) (0.27) Distance ( 5.15 (5.35 Ϫ0.19 0.71 (0.38) (0.34) (0.51) N (45 (58

identiªcation To exploit the randomness in the success and failure of decapitation attempts for causal identiªcation, my estimation strategy was to use simple OLS (ordi- nary least squares) regressions that take the following form:

Yi ϭ␤0 ϩ␤1SUCCESSi ϩ␥iXi ϩ εi,

where i indexes a campaign-year in which there is an decapitation attempt, Yi is the dependent variable (campaign termination, campaign outcome, conºict intensity, or insurgent attacks), SUCCESSi is a dummy variable equal to “1” if a leader is captured or killed in that campaign-year and “0” if the leader escapes any attempts, and Xi is a vector of other regressors. The key identiªcation assumption is that, conditional on observables, SUCCESSi is exogenous. Then, E|ε/SUCCESS,X] ϭ 0, and the average treatment effect is:

␤1 ϭ E[Y]SUCCESS ϭ 1,X] Ϫ E[Y]SUCCESS ϭ 0,X].

This equation shows that the estimates from the OLS regression equation written above identify the difference between successful and failed decapita-

once ªxed effects are added, population does not predict decapitation attempts. The results of this analysis are similar, supporting the plausibility of the exogeneity assumption.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/ISEC_a_00076 by guest on 26 September 2021 International Security 36:4 62

tion attempts. If the regressions reject that ␤ is “0,” then the outcomes of bids to decapitate insurgencies have a causal effect. Decapitation attempts do not occur randomly, so this empirical approach cannot conclusively demonstrate whether any observed effect is caused by success, failure, or both success and failure. Addressing this question requires a different approach. I used propensity-score matching to parse the effects of successful and unsuccessful leadership targeting. Matching is the preferred approach because it can ensure balance on observed covariates. Given that matching cannot ensure balance on unobserved covariates, however, these results are admittedly more speculative than those presented in the next section.

Results

This section presents the main results of my analysis. I begin by using OLS regressions with robust standard errors adjusted for clustering at the campaign level.40 All regressions include ªxed effects for the number of attempts that occurred in a campaign-year and for the decade during which each attempt occurred. I also include ªxed-effects speciªcations for the type of attempt (i.e., the method used to target the insurgent leader) and the region in which attempts occurred. These speciªcations are labeled in each table.

termination Table 3 displays estimates of leadership decapitation’s effect on campaign outcomes. The ªrst three columns estimate the effect of decapitation on war termination. The dependent variable is campaign termination; it is a dummy coded “1” if the campaign ended in the year in which a decapitation strike took place. The ªrst column presents the results without additional ªxed effects; the second column includes ªxed effects for attempt type; the third column includes ªxed effects for region; and the fourth column includes ªxed effects for both attempt type and region. In each regression, campaign termination was regressed on decapitation attempts’ success or failure; the results are estimates of the average effect of successful decapitation strikes compared with failed attempts. The results displayed in table 3 suggest that campaigns are more likely to end after leadership decapitation. The estimate shown in column 1 suggests

40. On using OLS to analyze limited-dependent variables such as those examined below, see Joshua D. Angrist, “Estimation of Limited-Dependent Variable Models with Dummy Endogenous Regressors: Simple Strategies for Empirical Practice,” Journal of Business and Economic Statistics, Vol. 19, No. 1 (January 2001), pp. 2–16. Probit regressions (not reported), however, produce similar results.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/ISEC_a_00076 by guest on 26 September 2021 Does Decapitation Work? 63

Table 3. Leadership Decapitation and Campaign Outcomes

Dependent Variable: Termination Dependent Variable: Victory (1) (2) (3) (4) (5) (6) Success 0.273*** 0.290*** 0.260*** 0.321*** 0.338*** 0.310*** (0.079) (0.081) (0.091) (0.073) (0.075) (0.084) Constant Ϫ0.140** Ϫ0.319** Ϫ0.427** Ϫ0.173** Ϫ0.416*** Ϫ0.505*** (0.068) (0.126) (0.179) (0.075) (0.129) (0.171) Type no yes yes no yes yes (ªxed effect) Region no no yes no no yes (ªxed effect) N 103 103 103 103 103 103

Robust standard errors appear in parentheses. ***p Ͻ 0.01, **p Ͻ 0.05, *p Ͻ 0.10.

that leadership decapitation increases the probability of war termination by 27 percentage points and is signiªcant at the 1 percent level. This result is robust: the estimates displayed in columns 2 and 3 range from 0.26 to 0.29, and all of the speciªcations are signiªcant at the 1 percent level. In each speciªca- tion, the lower bound of the 95 percent conªdence interval of the estimate is above “0.” These results change little when attempt type or region ªxed effects are included. They are also robust to nonparametric modeling; in each of the nonparametric speciªcations, the results are signiªcant at the 1 percent level.

victory Columns 4–6 of table 3 estimate the effect of decapitation on government victory. The dependent variable is a dummy coded “1” if the government defeated its opponent in the year that the decapitation strike occurred. As in the termination analysis, the independent variable is attempt success or failure, which is a dummy coded “1” if an attempt succeeded at capturing or killing its target. The results suggest that governments are more likely to win when they successfully target militant leaders. The coefªcient in column 4 shows that in- cumbents were an estimated 32 percentage points more likely to defeat insurgencies in years where counterinsurgents removed militant leaders than in years where such attempts failed—a sizable advantage. This estimate is statistically signiªcant at the 1 percent level. Like the results reported in the campaign termination analysis, the effect of decapitation victory is robust. In each

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/ISEC_a_00076 by guest on 26 September 2021 International Security 36:4 64

speciªcation, decapitation is signiªcant at the 1 percent level; the lower-bound estimate of the 95 percent conªdence interval is above zero; and various com- binations of ªxed effects do not change the effect. In sum, the evidence strongly suggests that leadership decapitation has substantial causal effects on campaign outcomes—removing militant leaders increases counterinsurgents’ chances of achieving quick, successful campaign terminations.

lethality Table 4 displays the estimated causal effect of leadership decapitation on the intensity of violence in counterinsurgency. Consistent with the evidence that leadership decapitation increases the likelihood of war termination and counterinsurgent victory, the evidence also suggests that leadership decapitation reduces conºict violence. Columns 1 and 2 in table 4 show the results of negative binomial regressions in which the dependent variable is the number of people killed by an insurgency in a given campaign-year. As expected, the sign of the point estimate in column 1 is negative, but it is not statistically sig- niªcant. The speciªcation in column 2 estimates the same negative binomial model, but it adds region and attempt type ªxed effects. This regression is signiªcant at the 1 percent level, and the point estimate is more than twice the size of the coefªcient in column 1, suggesting that once differences that owe to attempt type and region are accounted for, leadership decapitation is associated with decreased violence.41 The result in column 3 provides additional conªrmatory evidence. It presents the same negative binomial regressions as columns 1 and 2 but includes a lag of the dependent variable on the right-hand side of the equation. Accounting for the number of people killed by insurgencies at time t Ϫ 1 helps control for any unobserved cross-sectional differences in insurgent violence. The estimated effect of decapitation is again negative and signiªcant at the 1 percent level.

attacks Next, I examined the impact of leadership decapitation on the rate of insurgent attacks. It is useful to examine attacks separately from lethality because

41. As before, all regressions include ªxed effects for the decade in which each decapitation attempt occurred and for the number of attempts that were carried out during each campaign-year with at least one attempt. This ensures that any violence-reducing effect observed in the regression results cannot be attributed to unusually aggressive targeting operations or temporal trends. For example, during the Cold War, there was less stigma attached to targeted killing programs that aimed to eliminate enemy leaders in covert operations, a trend which was at least temporarily re- versed during the 1990s.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/ISEC_a_00076 by guest on 26 September 2021 Does Decapitation Work? 65

Table 4. Leadership Decapitation and Violence

Dependent Variable: Conºict Intensity Dependent Variable: Attacks (1) (2) (3) (4) (5) (6) Success Ϫ0.774 Ϫ1.994*** Ϫ1.637*** 0.212 Ϫ0.728** Ϫ1.685*** (0.494) (0.426) (0.420) (0.480) (0.328) (0.444) Constant Ϫ1.677* Ϫ1.640* Ϫ1.308 Ϫ0.925* Ϫ1.420* Ϫ1.392* (0.890) (0.868) (0.950) (0.552) (0.860) (0.811) Type no yes yes no yes yes (ªxed effect) Region no yes yes no yes yes (ªxed effect) Lagged no no yes no no yes dependent variable N 102 102 90 102 102 90

Robust standard errors appear in parentheses. ***p Ͻ 0.01, **p Ͻ 0.05, *p Ͻ 0.10.

whereas violence usually captures the quality of militant operations—that is, the extent to which the insurgency is able to inºict losses on its targets—attack frequency captures the pace of militant activities. Militants using a guerrilla warfare strategy might conduct many small attacks, with lower levels of lethality, in order to harass and intimidate its adversaries, whereas other militant organizations may conduct “spectacular” but infrequent attacks.42 I use a similar approach to estimate the effect of decapitation on insurgent attacks. The baseline speciªcations are negative binomial regressions, speciªed in table 4 both without (column 4) and with (column 5) ªxed effects. A lagged dependent variable speciªcation is shown in column 6. The results suggest that, on average, decapitation is associated with fewer insurgent attacks. The results of the ªxed-effects speciªcations in columns 5 and 6, for example, are negative and statistically signiªcant at the 5 and 1 percent levels, respectively. The coefªcient shown in column 4 is also negative, but it is small and insig- niªcant. As with the violence analysis, this suggests that, after including basic controls, decapitation is associated with fewer insurgent attacks.

42. Indeed, scholars have also suggested that decapitation has different effects on levels of militant lethality than it does on rates of militant attacks. For example, Byman observed that Israel’s targeted killing program reduced the lethality of Hamas attacks but increased the total number of attacks. Byman, “Do Targeted Killings Work?” p. 103.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/ISEC_a_00076 by guest on 26 September 2021 International Security 36:4 66

success and failure Targeting insurgent leaders is a game of chance. More bids to capture or kill insurgent leaders fail than succeed. But what are the consequences of failure? The element of chance in leadership targeting enables causal identiªcat- ion. The evidence presented above suggests that these outcomes—successful versus failed decapitation strikes—have a signiªcant impact on the dynamics and outcomes of counterinsurgency campaigns. Although it is tempting to attribute the causal effects identiªed above to successful leadership targeting, doing so would be premature. Because my identiªcation strategy does not include an untreated control group—it instead uses exogenous variation in successful and failed attempts—the possibility that failed attempts are driving the observed relationships cannot be ruled out. Speciªcally, negative externalities of failed decapitation strikes could underlie the identiªed effects. That is, when the tactics commonly used in decapitation strikes, such as bombings and raids, fail to eliminate insurgent leaders but incite mass resentment, these failures could decrease the chances of war termination and counterinsurgent victory and increase the chances of escalated levels of insurgent violence. To account for this potential issue, I assessed the impacts of both successful and failed decapitation strikes on counterinsurgency outcomes. Identifying the independent effects of success and failure alone is more difªcult than identifying the difference between them because, while the evidence suggests that the success of decapitation strikes is exogenous, conditional on an attempt taking place, decapitation strikes themselves do not occur at random. Treating them as if they do risks conºating the effects of successful and failed decapitation strikes with change that would have occurred regardless. For example, if decapitation strikes are more likely to occur when counterinsurgents believe that insurgents are growing, or are going to grow, in strength and lethality, they have an incentive to strike at the insurgency’s leadership from a position of relative weakness. This dynamic appears to be motivating the United States’ escalation of high-value targeting in Pakistan’s Federally Administered Tribal Areas, for example. These strategic choices could lead analysts to erroneously attribute any observed growth in militancy to high-value targeting attempts, even though high-value targeting might have had no impact on the insurgency’s capabilities or behavior.43

43. Alexander B. Downes makes an analogous argument about states’ decisions to use force against civilians during desperate moments of interstate wars. See Downes, “Desperate Times, Desperate Measures: The Causes of Civilian Victimization in War,” International Security, Vol. 30, No. 4 (Spring 2006), pp. 152–195. For a general theory of leaders’ decisions to “gamble for resurrec- tion” in war, see H.E. Goemans, War and Punishment: The Causes of War Termination and the First World War (Princeton, N.J.: Princeton University Press, 2000).

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/ISEC_a_00076 by guest on 26 September 2021 Does Decapitation Work? 67

Propensity-score matching is the best tool available for addressing this challenge. Propensity-score matching uses observable characteristics to predict decapitation strikes and then uses this information to stratify the sample into control and treatment groups that are balanced on observables.44 Stratifying the sample in this way enables me to compare similar years with and without decapitation strikes as if they were similar treatment and control groups.45 To implement this approach, for all countries i engaged in counterinsurgency campaigns in all years t, I used the following equation:

P(ATTEMPTit) ϭ⌽(Xit).

This estimator is used to predict attempts conditional on observables. Based on the predicted probabilities estimated from this equation, I formed four blocks, denoted by b, for varying levels of the propensity score and then checked the balance on the treatment and control covariates in each block. Once the sample was stratiªed and balanced, I estimated regressions using the following equation:

Yib ϭ␣b ϩ␤1SUCCESSib ϩ␤2FAILUREib ϩ␥Xib ϩ εib,

46 where ␣b indicates ªxed effects for each propensity score block. Table 5 shows separate estimates of the effects of successful and failed decapitation attempts on campaign termination and success. For each of these dependent variables, three speciªcations are presented: column 1 presents an OLS regression without controls, ªxed effects, or matching; the speciªcation in

44. Daniel E. Ho, Kosuke Imai, Gary King, and Elizabeth A. Stuart, “Matching as Nonparametric Preprocessing for Reducing Model Dependence in Parametric Causal Inference,” Political Analysis, Vol. 15, No. 3 (January 2007), pp. 199–236; and Donald B. Rubin, Matched Sampling for Causal Effects (New York: Cambridge University Press, 2006). 45. Compared with experimental and quasi-experimental approaches, the major weakness of matching is that assignment to treated and control groups is based on observables. Without full knowledge of the data-generating process, it remains possible that my estimates will be biased because of selection on unobservables. Although selection bias is a potential concern when using any matching estimator, matching remains a useful technique for scholars seeking to identify causal effects because it ensures sample balance on observables and excludes extreme counterfactuals. See Gary King and Langche Zeng, “The Dangers of Extreme Counterfactuals,” Political Analysis, Vol. 14, No. 2 (November 2006), pp. 131–159. 46. Before estimating the effects of successful and failed decapitation strikes, I examined whether pretreatment covariates in my dataset predicted observed decapitation attempts. Supplemental table 4 in the online appendix shows the results of probit regressions that included the same variables used in the t-tests presented in table 2 above. The incumbent’s GDPPC is a signiªcant predictor of decapitation attempts in two of six speciªcations. This is intuitive: wealthier governments should be more likely to have the capability to aggressively target insurgent leaders. I control for this variable in the regressions.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/ISEC_a_00076 by guest on 26 September 2021 International Security 36:4 68

Table 5. The Impact of Successful versus Failed Attempts

Dependent Variable: Conºict Intensity Dependent Variable: Attacks (1) (2) (3) (4) (5) (6) Success 0.282*** 0.284*** 0.284*** 0.296*** 0.289*** 0.288*** (0.072) (0.073) (0.074) (0.069) (0.070) (0.070) Failure Ϫ0.010*** Ϫ0.022 Ϫ0.030 Ϫ0.016 Ϫ0.022 Ϫ0.021 (0.037) (0.037) (0.033) (0.027) (0.028) (0.028) Constant 0.039 0.155 0.028 0.020 0.012 Ϫ0.220 (0.032) (0.112) (0.250) (0.027) (0.070) (0.193) Controls no yes yes no yes yes Matching no no yes no no yes N 932 932 932 932 932 932

Robust standard errors appear in parentheses. ***p Ͻ 0.01, **p Ͻ 0.05, *p Ͻ 0.10.

column 2 includes control variables and ªxed effects for attempt type and region; and column 3 includes all of these regressors and uses matching to ensure balance on covariates. The results strongly suggest that successful leadership removals, not blowback from failed attempts, drive the overall effect of leadership decapitation. The results are displayed in table 5. For both successful and unsuccessful decapitation attempts, the results are similar across the matched and unmatched samples and are robust to the inclusion of ªxed effects and controls. Con- ªdence in the ªndings is enhanced by the fact that the point estimates for successful leadership removals shown in table 5 are consistent with those displayed in table 3, which used exogenous variation in attempt outcomes to examine whether the correlation between successful attempts and campaign termination and government victory differed from the correlations between failed attempts and those outcomes. Speciªcally, the estimated effect of successful decapitation attempts remains similar in size to the estimates shown in table 3. The coefªcients presented in table 5 suggest that leadership decapitation is associated with a 28-percentage point increase in the probability of termination during the year in which the decapitation attempt occurred and a 29- or 30-percentage point increase in the probability of government victory. Both of these results are signiªcant at the 1 percent level. There is little evidence of a “blowback effect.” As the blowback hypothesis would predict, the point estimates for failed attempts are negative, which indicates that failed attempts to capture or kill insurgent leaders may have counterproductive effects on governments’ chances of defeating insurgencies.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/ISEC_a_00076 by guest on 26 September 2021 Does Decapitation Work? 69

There is not enough evidence, however, to reject the null hypothesis—that failed decapitation operations have no overall impact on states’ chances of strategic success. Indeed, the estimated effect of failed attempts is small and far from statistically signiªcant, with p-values that range from 0.356 to 0.788. Taken together, this evidence strongly indicates that the successful removal of insurgent leaders, not blowback from failed attempts, underlies my key ªnd- ings on the effects of leadership decapitation in counterinsurgency operations.

Results II: When Is Decapitation Effective?

The results presented above strongly suggest that decapitating insurgencies has causal effects on counterinsurgency campaign termination, outcomes, levels of conºict intensity, and insurgent attack rates. Another key question is whether leadership decapitation is more or less effective against some types of insurgencies than others. I examine this question below.

ideological and identity-based insurgencies A central debate in the civil war literature concerns whether the dynamics of “ideological” and “identity” conºicts differ.47 Communist insurgencies are usually categorized as ideological movements; ethnic and religious groups are usually classiªed as identity-based movements. Scholars suggest that a key distinction between ideological and identity-based conºicts is that ideological conºicts are fought over how polities should be governed, whereas identity conºicts usually involve at least one party that views itself as fundamen- tally different from the other and, consequently, is ªghting to pursue some form of self-determination. This argument implies that fundamental differences between belligerent parties characterize identity conºicts, making these wars more difªcult to resolve.48 Ideology is more malleable, scholars argue, and so ideological conºicts are easier to resolve.49

47. Chaim Kaufmann, “Possible and Impossible Solutions to Ethnic Civil Wars,” International Se- curity, Vol. 20, No. 4 (Spring 1996), pp. 136–175; and Nicholas Sambanis, “Partition as a Solution to Ethnic War: An Empirical Critique of the Theoretical Literature,” World Politics, Vol. 52, No. 4 (July 2000), pp. 437–483. 48. Monica Duffy Toft, The Geography of Ethnic Violence: Identity, Interests, and the Indivisibility of Ter- ritory (Princeton, N.J.: Princeton University Press, 2003); Stacie E. Goddard, “Uncommon Ground: Indivisible Territory and the Politics of Legitimacy,” International Organization, Vol. 60, No. 1 (Win- ter 2006), pp. 35–68; Ron E. Hassner, “The Path to Intractability: Time and the Entrenchment of Ter- ritorial Disputes,” International Security, Vol. 31, No. 3 (2006/07), pp. 107–138; and Monica Duffy Toft, “Getting Religion? The Puzzling Case of Islam and Civil War,” International Security, Vol. 31, No. 4 (Spring 2007), pp. 97–131. 49. On the challenges of resolving ethnic conflicts via negotiated settlement, see Alexander B. Downes, “The Problem with Negotiated Settlements to Ethnic Civil Wars,” Security Studies, Vol. 13,

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/ISEC_a_00076 by guest on 26 September 2021 International Security 36:4 70

Table 6. Leadership Decapitation and Ideological Insurgencies

Dependent variable: Victory (1) (2) (3) (4)

Success 0.361*** 0.367*** 0.327*** 0.333*** (0.086) (0.083) (0.094) (0.093) Ideological 0.140 0.139 0.095 0.064 (0.102) (0.106) (0.084) (0.100) Success*Ideological 0.143 0.124 Ϫ0.109 Ϫ0.086 (0.183) (0.176) (0.178) (0.180) Constant Ϫ0.222*** Ϫ0.433*** Ϫ0.312** Ϫ0.472** (0.075) (0.144) (0.138) (0.202) Type (ªxed effects) no yes no yes Region (ªxed effects) no no yes yes Observations 102 102 102 102

Robust standard errors appear in parentheses. ***p Ͻ 0.01, **p Ͻ0 .05, *p Ͻ 0.10.

To the extent that this argument is true, an observable implication is that ideological insurgencies should be more likely to collapse after a leadership decapitation incident because these groups should, on average, have less resolve against a demonstrably threatening government or third-party counterinsurgent than groups engaged in identity-based conºicts. Yet scholars have made the opposite claim: they argue that insurgencies with a communist ideology tend to be particularly durable because of their mass-based ideology and political organizations, which are typically inºu- enced by Soviet political organization.50 The implication of this argument is that communist insurgencies’ durability should make them more resilient to the loss of key individual leaders than other types of insurgencies. I examine these competing hypotheses empirically in tables 6 and 7; the results support neither argument. The results are from OLS regressions in which counterinsurgency victory is regressed on the outcome of decapitation attempts against communist insurgent leaders.51 The main effect of successful

No. 4 (Summer 2004), pp. 230–279; and on religion, see Toft, “Getting Religion?”; and Ron E. Hassner, War on Sacred Grounds (Ithaca, N.Y.: Cornell University Press, 2009). 50. Lucian Pye, Guerrilla Communism in Malaya (Princeton, N.J.: Princeton University Press, 1956); and Weinstein, Inside Rebellion, pp. 20–31. 51. Because these are saturated regression models, or regressions with discrete explanatory variables where the model includes a separate parameter for all possible values taken by the explanatory variables, the conditional expectation function is fit correctly regardless of the distribution of Yi. Linear models can thus be used to estimate the effects of the regressors. OLS regression has numerous advantages over nonlinear models in this situation, because nonlinear models require the researcher to make numerous decisions (e.g., the weighting scheme and derivatives vs. finite dif-

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/ISEC_a_00076 by guest on 26 September 2021 Does Decapitation Work? 71

Table 7. Leadership Decapitation and Identity-Based Insurgencies

Dependent variable: Victory (1) (2) (3) (4)

Success 0.317*** 0.350*** 0.292*** 0.330*** (0.100) (0.105) (0.107) (0.114) Identity 0.050 0.115* Ϫ0.032 0.098 (0.046) (0.066) (0.112) (0.114) Success*Identity 0.021 Ϫ0.017 0.026 Ϫ0.014 (0.155) (0.145) (0.161) (0.153) Constant Ϫ0.213*** Ϫ0.452*** Ϫ0.264 Ϫ0.560*** (0.073) (0.121) (0.166) (0.204) Type (ªxed effects) no yes no yes Region (ªxed effects) no no yes yes Observations 102 102 102 102

Robust standard errors appear in parentheses. ***p Ͻ 0.01, **p Ͻ0 .05, *p Ͻ 0.10.

decapitation strikes is statistically signiªcant at the 1 percent level across the speciªcations. The preferred speciªcation, which includes the full battery of ªxed effects and is shown in column 4, yields a point estimate of 0.333, with a standard error of 0.093. The point estimate for successful attempts, conditional on the target being a communist leader, is negative across the regressions shown in table 8, but it is not statistically signiªcant in any of the speciªcations. It is therefore possible that the effect of leadership decapitation on communist insurgencies is negative, but there is insufªcient evidence to demonstrate this conclusively. Similarly, the results for identity-based groups, which are given in table 8, show that leadership decapitation’s main effect is positive, but the effect of decapitating identity-based insurgencies, which I proxy using insurgencies with secessionist aims, does not differ signiªcantly from that of other types of insurgencies. No clear pattern can be discerned from the interaction term. The effect of decapitating identity-based insurgencies appears to be small and the direction of the relationship dependent on model speciªcation. Thus it does not appear that the effect of removing the leaders of identity-

ferences), whereas OLS is standardized and yields estimates that are consistent with the marginal effects that must be computed from the estimates of nonlinear models in order to produce a causal interpretation of these estimates. For more on this methodological choice, see Joshua D. Angrist, “Estimation of Limited-Dependent Variable Models with Dummy Endogenous Regressors,” Jour- nal of Business & Economic Statistics, Vol. 19, No. 1 (January 2001), pp. 2–28; and Joshua D. Angrist and Jörn-Steffan Pischke, Mostly Harmless Econometrics: An Empiricist’s Companion (Princeton, N.J.: Princeton University Press, 2009), pp. 48–51, 102–107.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/ISEC_a_00076 by guest on 26 September 2021 International Security 36:4 72

Table 8. Leadership Decapitation and Center-Seeking Insurgencies

Dependent variable: Victory (1) (2) (3) (4)

Success 0.354*** 0.354*** 0.354*** 0.354*** (0.111) (0.111) (0.111) (0.111) Center-Seeking Ϫ0.037 Ϫ0.037 Ϫ0.037 Ϫ0.037 (0.038) (0.038) (0.038) (0.038) Success*Center-Seeking Ϫ0.069 Ϫ0.069 Ϫ0.069 Ϫ0.069 (0.144) (0.144) (0.144) (0.144) Constant 0.037 0.037 0.037 0.037 (0.038) (0.038) (0.038) (0.038) Type (ªxed effects) no yes no yes Region (ªxed effects) no no yes yes Observations 102 102 102 102

Robust standard errors appear in parentheses. ***p Ͻ 0.01, **p Ͻ0 .05, *p Ͻ 0.10.

based insurgent groups differs from the effects of removing the leaders of other types of insurgencies. And overall, the analysis reveals no evidence that the distinction between ideologically and identity-based conºicts is salient in terms of leadership decapitation’s effectiveness.

center-seeking insurgencies Having examined the effect of decapitation on secessionist insurgencies, I now turn to center-seeking insurgencies. Center-seeking insurgencies are those guerrilla movements whose stated goal is to become the sovereign government of the countries in which they ªght, regardless of their ideology or organization. Although all civil wars are difªcult to resolve through negotiations, conºicts in which insurgencies seek the center are particularly difªcult to me- diate. James Fearon and David Laitin show that conºicts with center-seeking insurgencies are less likely to end in negotiated settlement than are secessionist conºicts, possibly as a result of commitment problems.52 If the nature of insurgencies’ aims affects the parties’ ability to credibly commit to peace, one might expect that decapitating center-seeking insurgencies would make it even more difªcult for governments to credibly commit, reducing the chances of a successful resolution. I test this proposition in table 8. The results displayed in table 8 show that although the point estimates of the interaction of Success*Center-Seeking are all in the expected negative di-

52. Fearon and Laitin, “Civil War Termination.”

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/ISEC_a_00076 by guest on 26 September 2021 Does Decapitation Work? 73

Table 9. Leadership Decapitation and Islamist Insurgencies

Dependent variable: Victory (1) (2) (3) (4)

Success 0.316*** 0.336*** 0.268*** 0.283*** (0.087) (0.088) (0.089) (0.091) Islamist Ϫ0.027 Ϫ0.030 Ϫ0.008 Ϫ0.063 (0.039) (0.053) (0.118) (0.148) Success*Islamist 0.063 0.043 0.154 0.134 (0.186) (0.170) (0.216) (0.203) Constant Ϫ0.168** Ϫ0.357*** Ϫ0.284** Ϫ0.430** (0.081) (0.123) (0.120) (0.167) Type (ªxed effects) no yes no yes Region (ªxed effects) no no yes yes Observations 102 102 102 102

Robust standard errors appear in parentheses. ***p Ͻ 0.01, **p Ͻ0 .05, *p Ͻ 0.10.

rection, they are small and statistically insigniªcant. Thus, although it remains possible that decapitation is systematically less effective against center-seeking insurgencies, there is little evidence to substantiate such a claim.

islamist insurgencies Analysts have suggested that leadership decapitation is not an effective instru- ment against Islamist insurgencies—perhaps because several recent high- proªle Islamist groups, such as al-Qaida in Iraq, have operated through relatively ºat, cell-based organizations. This proposition is examined in table 9. The interaction between Success*Islamist shows a relatively small and insigniªcant relationship between successful decapitation attempts against Islamist insurgencies and counterinsurgency victory. Detecting a statistical effect is unlikely, however, because the number of decapitations of Islamist insurgent leaders in the data set is small (N ϭ 10). Yet the consistently positive relationship between Success*Islamist and government victory suggests it is possible that using decapitation strategies against Islamist groups is more effective than previously believed. And, to the extent that Islamist groups orga- nize as ºat, cell-based networks, removing the leaders of such decentralized organizations may be an effective means of exploiting organizational vulnera- bilities. Given that my dataset does not include Islamist insurgencies from the post–September 11, 2001, period, however, caution should be used in extrapo- lating from these ªndings.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/ISEC_a_00076 by guest on 26 September 2021 International Security 36:4 74

Table 10. Leadership Decapitation and Organizational Age

(1) (2) (3) (4) Dependent variable: Victory 0–10 Year Campaigns 11ϩ Year Campaigns

Success (0.279*** (0.279*** (0.455** (0.455** (0.083) (0.083) (0.156) (0.156) Constant (0.0238 (0.0238 (0 (0 (0.024) (0.024) (0.001) (0.001) Type (ªxed effects) no yes no yes Region (ªxed effects) no yes no yes Observations 75 75 27 27

Robust standard errors appear in parentheses. ***p Ͻ 0.01, **p Ͻ0 .05, *p Ͻ 0.10.

organizational age A key argument in Jordan’s article on the effectiveness of decapitating terrorist organizations is that decapitation is particularly ineffective against older organizations. She claims that older terrorist organizations are more resilient to decapitation, both when compared with younger organizations that suffered decapitation and with similarly aged organizations that never suffered decapitation.53 I test this claim indirectly by examining the impact of leadership decapitation during different time periods of counterinsurgency campaigns to assess whether insurgencies that are strong enough to survive to ªght longer campaigns are also more resilient to leadership decapitation. The results are displayed in table 10. Table 10 compares the outcomes that follow successful and failed decapitation attempts in the ªrst ten years of counterinsurgency campaigns as well as attempts that occur in campaigns lasting longer than ten years. Essentially, this compares attempts in the upper quartile of attempts, by duration, with attempts in the bottom three quartiles. The ten-year length is also a substantively recognizable duration, now that the U.S. campaign in Afghanistan is in its eleventh year, and key questions about the effects of U.S. high-value targeting loom as the U.S. government searches for a way to wind down the war on ac- ceptable terms.54

53. Jordan, “When Heads Roll,” pp. 740–747. For more on this hypothesis, see Martha Crenshaw, “How Terrorism Declines,” Terrorism and Political Violence, Vol. 3, No. 1 (March 1991), p. 79. Mi- chael C. Horowitz also uses organizational age to explain terrorist group activities. See Horowitz, The Diffusion of Military Power: Causes and Consequences for International Politics (Princeton, N.J.: Princeton University Press, 2010). 54. It should be reiterated that these results are not directly applicable for predicting the effect of

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/ISEC_a_00076 by guest on 26 September 2021 Does Decapitation Work? 75

The point estimates of success are in the expected positive direction and are statistically signiªcant at the 5 percent level or above, suggesting that governments are more likely to defeat insurgencies after successful decapitations than after failed attempts regardless of the campaign’s duration. Interestingly, these results suggest that the impact of removing insurgent leaders might be even larger in long campaigns than in shorter campaigns.55 This is consistent with the hypothesis that leadership decapitation can help break the morale of insurgencies that have been engaged in long, often difªcult, campaigns.

capture versus killing Finally, I examine the relationships between the methods of targeting and war outcomes. Several scholars have made a distinction between capture versus killing, suggesting that capture should be more effective because it can lead to intelligence gains through interrogation.56 To test this hypothesis, I categorized targeting attempts either as “lethal” or “nonlethal” based on my coding of attempt types. Bombings, combat, and shootings were all categorized as lethal attempts; raids and sweep operations, which can be lethal but often are not in- tended to be, were categorized as nonlethal unless they ultimately resulted in the use of lethal force.57 Twenty-ªve nonlethal attempts are documented in the data, as compared to ninety-four attempts categorized as lethal. Contrary to this conventional wisdom, the results shown in table 11 suggest that killing insurgent leaders is likely to be more effective than capturing them. There is a statistically signiªcant relationship between operations that resulted in the deaths of militant leaders and quicker war termination and government victory. This result remains similar when I control for failed attempts and other covariates, and it is robust to the matching estimator described above. Opera-

successful high-value targeting in Afghanistan or Pakistan. Rather, they provide general information on the average effect of leadership decapitation in counterinsurgency campaigns across a large number of cases since the 1970s. 55. To be sure, although these findings are statistically significant, given the relatively small sample of long wars (N = 27), the precision of the point estimates is likely to be reduced. The results (not reported) are similar, however, when other duration cut points are used. 56. See Frankel, “The ABCs of HVT,” pp. 24–26; and Cronin, “How al-Qaida Ends,” p. 736. Jor- dan’s research suggests that capturing “upper echelon” terrorists (as opposed to terrorist leaders) might be more strongly associated with terrorist organization collapse than killing upper echelon members, but that there is a stronger association between killing top leaders and organizational collapse than between capturing them and organizational collapse. See Jordan, “When Heads Roll,” p. 736. 57. In his June 2011 testimony to the Senate Armed Services Committee, for example, commander of the Joint Special Operations (JSOC) Command, Vice Adm. William McRaven, disclosed that JSOC commandos in Afghanistan had discharged their firearms in only about 15 percent of the raids conducted in 2010. See Aamer Madhani, “Special Ops Nominee Defends Night Raids in Afghan War,” National Journal, June 28, 2011, http://www.nationaljournal.com/nationalsecurity/ special-ops-nominee-defends-night-raids-in-afghan-war-20110628.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/ISEC_a_00076 by guest on 26 September 2021 International Security 36:4 76

Table 11. Capture versus Killing

(1) (2) (3) (4) Dependent variable Termination Termination Victory Victory

Kill 0.382*** 0.378*** 0.367*** 0.369*** (0.094) (0.094) (0.093) (0.093) Capture 0.025 0.028 0.069 0.066 (0.095) (0.095) (0.094) (0.094) Failure Ϫ0.004 Ϫ0.008 Ϫ0.004 Ϫ0.001 (0.037) (0.036) (0.028) (0.028) Constant 0.097 0.234 Ϫ0.029 Ϫ0.115 (0.109) (0.172) (0.068) (0.124) Controls yes yes yes yes Matching no yes no yes Observations 927 927 927 927

Robust standard errors appear in parentheses. ***p Ͻ 0.01, **p Ͻ0 .05, *p Ͻ 0.10.

tions that resulted in the capture of militant leaders are also positively associated with these outcomes, but these results are smaller and statistically insigniªcant. These results are consistent with Jordan’s ªnding that killing top leaders is more frequently associated with successful countermilitancy than ar- resting them.58 More important, they not only suggest that killing militant leaders represents a severe blow to militant organizational capacity and morale, but they are consistent with arguments made by U.S. counterterrorism strategists that high-value targeting operations themselves can yield tangible information through sensitive site-exploitation techniques and other activities that may lead to additional operations. These claims were substantiated by the intelligence windfall that resulted from the Navy SEAL raid that killed Osama bin Laden in May 2011.59

Conclusion

Targeting militant leaders is now a centerpiece of U.S. strategy in Afghanistan, Pakistan, Somalia, and Yemen, but does capturing or killing militant leaders

58. Jordan, “When Heads Roll,” pp. 736. 59. On targeting and intelligence collection, see Michael T. Flynn, Rich Juergens, and Thomas Cantrell, “Employing ISR: SOF Best Practices,” Joint Forces Quarterly, Vol. 50, No. 3 (Fall 2008), pp. 57–61. On the intelligence collected during the bin Laden raid, see, for example, Andrew Tilghman, “Bin Laden Raid Also Yields Trove of Intel,” Army Times, May 2, 2011, http:// www.armytimes.com/news/2011/05/military-bin-laden-raid-intelligence-materials-050211w/.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/ISEC_a_00076 by guest on 26 September 2021 Does Decapitation Work? 77

work? Most extant research eschews the notion that removing enemy leaders can help governments achieve military and political goals. Regardless of whether a government’s adversary is a state, a terrorist organization, or a guerrilla insurgency, the scholarly opinion has been that high-value targeting is ineffective at best and counterproductive at worst. The evidence presented in this article challenges this view. In previous studies, causal inference and generalizability have been difªcult given research design and measurement issues. I addressed these issues by employing a data- driven approach in which I analyzed variation in the consequences of successful and failed decapitation attempts in campaigns dating from the mid-1970s. After correcting for the endogeneity and measurement issues that have hin- dered previous studies, I found that neutralizing insurgent leaders has a substantively large and statistically signiªcant effect on numerous metrics of countermilitancy effectiveness. Speciªcally, the results showed that removing insurgent leaders increases governments’ chances of defeating insurgencies, reduces insurgent attacks, and diminishes overall levels of violence. Because these effects were estimated by comparing the consequences of successful and failed decapitation attempts, I conducted additional analysis to ensure that the observed effects can be attributed to successful operations against insurgent leaders rather than to blowback from botched high-value targeting missions. This was conªrmed to be the case. Yet the data also show conclusively that killing or capturing insurgent leaders is usually not a silver bullet. Neutralizing insurgent leaders signiªcantly increases governments’ chances of reducing violence, terminating wars, and defeating insurgencies. A variety of different empirical tests consistently dem- onstrated that governments were more likely to defeat insurgencies following the successful removal of top insurgent leaders, but this probability was consistently estimated at around 25 to 30 percent—a far cry from the silver bullet many look for when they analyze leadership decapitation. Yet this effect indeed provides a sizable advantage, which can help explain why governments continue to invest in high-value targeting despite its legal ambiguity and nor- mative disrepute. These are not the only ªndings with policy implications. Importantly, the results do not support the common argument that the costs of failed targeting outweigh the beneªts of successful targeting; although there is abundant evidence that capturing or killing insurgent leaders is associated with key metrics of successful counterinsurgency, there is no credible evidence of a martyrdom effect, whereby trying but failing to neutralize militant leaders decreases governments’ chances of defeating insurgencies or increases levels of antigovern- ment violence. The apparently low costs of failed targeting to operational

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/ISEC_a_00076 by guest on 26 September 2021 International Security 36:4 78

effectiveness is consistent with choices made by states, such as Israel and the United States, to continue to aggressively target individual members of insurgent and terrorist organizations—including midlevel operatives who can potentially lead them to senior leaders—despite the inherent uncertainty, difªculty, and risks of doing so, and to continue to invest in intelligence capabilities and Special Operations Forces dedicated to kinetic and nonkinetic targeting. The role, responsibilities, and budget of the U.S. Special Operations Command continue to expand even as signiªcant budget cuts become a reality for the Department of Defense. As long as the United States continues to move its ªght from the battleªeld to the shadows, this trend will likely remain true.

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/ISEC_a_00076 by guest on 26 September 2021 Does Decapitation Work? 79

Appendix. Successful Leadership Decapitations

Government Insurgency Leader Year

Morocco Polisario El-Ouali Mustapha Sayed 1976 Indonesia Fretilin Nicolau Lobato 1978 Mozambique Renamo Andre Matsangaissa 1979 Nigeria Maitatsines Mohammadu Marwa 1980 India PLA N. Bisheswar Singh 1981 India PLA Thoundam Kunjabehari 1982 Colombia M-19 Carlos Toledo Plata 1984 Somalia SSDF Abdullahi Yusuf Ahmed 1984 India PREPAK R.K. Tulachandra 1985 Yemen YSP Abdul Fattah Ismail 1986 Pakistan MQM Altaf Hussain 1986 India KCF Manbir Singh Chaheru 1986 India KLF Aroor Singh 1988 India KLM Avtar Singh Brahma 1988 India KCF Labh Singh 1988 Sri Lanka JVP Rohana Wijeweera 1989 Pakistan MQM Altaf Hussain 1991 Peru Shining Path Abimael Guzmán 1992 Chad MDD Goukouni Guet 1992 Algeria GIA Mansouri Meliani 1992 India BKI Sukhdev Singh 1992 Indonesia Fretilin Xanana Gusmao 1992 India KLF Gurjant Singh Budhsinghwala 1992 Chad CNR Abbas Koty 1993 Algeria GIA Abdelhak Layada 1993 Indonesia Fretilin Antonio Gomes da Costa 1993 Algeria GIA Cherif Gousmi 1994 Algeria GIA Mourad Sid Ahmed 1994 India BTFK Gurbachan Singh Manochahal 1994 Russia Chechens Dzhokhar Dudayev 1996 Sierra Leone RUF Foday Sankoh 1997 Chad FDR Laokein Barde 1998 Philippines ASG Abdura jik Abubakar Janjalani 1998 Peru Shining Path Oscar Ramirez 1999 Angola FLEC-R Antonio Bento Bembe 1999 Turkey PKK Abdullah Ocalan 1999 Sierra Leone RUF Foday Sankoh 2000 India UNLF Samarendra Singh 2001 Chad MDJT Youssouf Togoimi 2002 Angola UNITA Jonas Savimbi 2002 Algeria GIA Antar Zouabri 2002 Algeria GSPC Nabil Sahraoui 2004 Russia Chechen Aslan Maskhadov 2005 United States/Iraq AQI Abu Musab al-Zarqawi 2006 Philippines ASG Khadaffy Janjalani 2006 Sri Lanka LTTE Velupillai Prabhakaran 2009

Downloaded from http://www.mitpressjournals.org/doi/pdf/10.1162/ISEC_a_00076 by guest on 26 September 2021